As humans, we do pretty well on guessing the genre of any song that we are listening to. As a part of our Data Mining & Analytics project, we tried to explore if we can train machines to do the same job well.

Using Python and Amazon Web Services, we built a system that takes in data from a collection of audio files of various genre and trains a recurrent neural network to recognize the genre of the input audio file. We were able to achieve an accuracy of 61% in predicting the genre of a given audio file. However, there was variance depending on the genre of the audio file. Classical music was almost always correctly classified. However, Jazz was frequently misclassified as metal or classical. Pop was misclassified by multiple models.

This project was a group effort. My fellow teammates Richa Prajapati, Keshav Potluri and Konstanin Lackner made this amazing initiative possible.